Asymmetric Goal Drift in Coding Agents Under Value Conflict

Magnus Saebo, Spencer Gibson, Tyler Crosse, Achyutha Menon, Eyon Jang, Diogo Cruz

Keywords: goal drift, AI safety, coding agents, value alignment, instruction following, adversarial pressure, model values, value hierarchies, agentic AI

TL;DR: Coding agents exhibit asymmetric goal drift, more readily violating system prompt constraints that oppose strongly-held values like security and privacy, and this drift is compounded by adversarial codebase comments and accumulated context.

Abstract:

Agentic coding agents are increasingly deployed autonomously, at scale, and over long-context horizons. Throughout an agent's lifetime, it must navigate tensions between explicit instructions, learned values, and environmental pressures, often in contexts unseen during training. Prior work on model preferences, agent behavior under value tensions, and goal drift has relied on static, synthetic settings that do not capture the complexity of real-world environments. To this end, we introduce a framework built on OpenCode to orchestrate realistic, multi-step coding tasks to measure how agents violate explicit constraints in their system prompt over time with and without environmental pressure toward competing values.
Using this framework, we demonstrate that GPT-5 mini, Haiku 4.5, and Grok Code Fast 1 exhibit asymmetric drift: they are more likely to violate their system prompt when its constraint opposes strongly-held values like security and privacy. We find for the models and values tested that goal drift depends on three compounding factors: value alignment, adversarial pressure, and accumulated context. However, even strongly-held values like security show non-zero violation rates under sustained environmental pressure. These findings reveal that shallow compliance checks are insufficient and that comment-based pressure can exploit model value hierarchies to override system prompt instructions. More broadly, our findings highlight a gap in current alignment approaches in ensuring that agentic systems appropriately balance explicit user constraints against broadly beneficial learned preferences under sustained environmental pressure.

Submission Number: 222